基于血管连通性的视网膜血管分割技术研究

Connection Sensitive Attention U-NET for Accurate Retinal Vessel Segmentation

为了提高智能医疗图像处理水平,减少人工分析病理图像,节约医院成本,提出一种基于血管连通性的视网膜血管分割技术,该技术采用连接敏感注意力U-Net(connection sensitive attention U-Net,CSAU)的网络模型,设计新型神经网络结构的注意门,结合提出的连接敏感损失,应用视网膜血管分割,从而实现对异常、分叉和微血管的血管分割。最后,通过在DRIVE、STARE和HRF数据集上进行测试,实验结果表明,基于血管连通性的视网膜血管分割技术可以通过对连通性和拓扑性进行建模保留细血管的连通性。

In order to improve the level of intelligent medical image processing,reduce manual analysis of pathological images,and save hospital costs.A retinal vascular segmentation technology based on vascular connectivity is proposed.This technology uses connection sensitive attention U-Net(CSAU),designing the attention gate of the new neural network structure,combining the proposed connection sensitivity loss,and applying retinal blood vessel segmentation to realize the blood vessel segmentation of abnormalities,bifurcations and micro-vessels.Finally,through testing on the DRIVE,STARE and HRF data sets,the experimental results show that the retinal vascular segmentation technology based on vascular connectivity can preserve the connectivity of small blood vessels by modeling connectivity and topology.